Charles Hong

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2024-237

December 20, 2024

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-237.pdf

In the hardware design space exploration process, it is critical to optimize both hardware parameters and algorithm-to-hardware mappings. Previous work has largely approached this simultaneous optimization problem by separately exploring the hardware design space and the mapspace---both individually large and highly nonconvex spaces---independently. The resulting combinatorial explosion has created significant difficulties for optimizers.

In this work, we introduce DOSA, which consists of differentiable performance models and a gradient descent-based optimization technique to simultaneously explore both spaces and identify high-performing design points. Experimental results demonstrate that DOSA outperforms random search and Bayesian optimization by 2.80x and 12.59x, respectively, in improving DNN model energy-delay product, given a similar number of samples. We also demonstrate the modularity and flexibility of DOSA by augmenting our analytical model with a learned model, allowing us to optimize buffer sizes and mappings of a real DNN accelerator and attain a 1.82x improvement in energy-delay product.

Advisors: Sophia Shao

---

BibTeX citation:

@mastersthesis{Hong:EECS-2024-237,
    Author= {Hong, Charles},
    Title= {Enhancing Accelerator Design Space Exploration with Differentiable Modeling and Unified Hardware-Software Co-Exploration},
    School= {EECS Department, University of California, Berkeley},
    Year= {2024},
    Month= {Dec},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-237.html},
    Number= {UCB/EECS-2024-237},
    Abstract= {In the hardware design space exploration process, it is critical to optimize both hardware parameters and algorithm-to-hardware mappings.
Previous work has largely approached this simultaneous optimization problem by separately exploring the hardware design space and the mapspace---both individually large and highly nonconvex spaces---independently. The resulting combinatorial explosion has created significant difficulties for optimizers. 

In this work, we introduce DOSA, which consists of differentiable performance models and a gradient descent-based optimization technique to simultaneously explore both spaces and identify high-performing design points. Experimental results demonstrate that DOSA outperforms random search and Bayesian optimization by 2.80x and 12.59x, respectively, in improving DNN model energy-delay product, given a similar number of samples. We also demonstrate the modularity and flexibility of DOSA by augmenting our analytical model with a learned model, allowing us to optimize buffer sizes and mappings of a real DNN accelerator and attain a 1.82x improvement in energy-delay product.},
}

EndNote citation:

%0 Thesis
%A Hong, Charles 
%T Enhancing Accelerator Design Space Exploration with Differentiable Modeling and Unified Hardware-Software Co-Exploration
%I EECS Department, University of California, Berkeley
%D 2024
%8 December 20
%@ UCB/EECS-2024-237
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-237.html
%F Hong:EECS-2024-237